1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 1995 Linus Torvalds
4 *
5 * This file contains the setup_arch() code, which handles the architecture-dependent
6 * parts of early kernel initialization.
7 */
8 #include <linux/acpi.h>
9 #include <linux/console.h>
10 #include <linux/crash_dump.h>
11 #include <linux/dma-map-ops.h>
12 #include <linux/dmi.h>
13 #include <linux/efi.h>
14 #include <linux/init_ohci1394_dma.h>
15 #include <linux/initrd.h>
16 #include <linux/iscsi_ibft.h>
17 #include <linux/memblock.h>
18 #include <linux/panic_notifier.h>
19 #include <linux/pci.h>
20 #include <linux/root_dev.h>
21 #include <linux/hugetlb.h>
22 #include <linux/tboot.h>
23 #include <linux/usb/xhci-dbgp.h>
24 #include <linux/static_call.h>
25 #include <linux/swiotlb.h>
26
27 #include <uapi/linux/mount.h>
28
29 #include <xen/xen.h>
30
31 #include <asm/apic.h>
32 #include <asm/numa.h>
33 #include <asm/bios_ebda.h>
34 #include <asm/bugs.h>
35 #include <asm/cpu.h>
36 #include <asm/efi.h>
37 #include <asm/gart.h>
38 #include <asm/hypervisor.h>
39 #include <asm/io_apic.h>
40 #include <asm/kasan.h>
41 #include <asm/kaslr.h>
42 #include <asm/mce.h>
43 #include <asm/mtrr.h>
44 #include <asm/realmode.h>
45 #include <asm/olpc_ofw.h>
46 #include <asm/pci-direct.h>
47 #include <asm/prom.h>
48 #include <asm/proto.h>
49 #include <asm/thermal.h>
50 #include <asm/unwind.h>
51 #include <asm/vsyscall.h>
52 #include <linux/vmalloc.h>
53
54 /*
55 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
56 * max_pfn_mapped: highest directly mapped pfn > 4 GB
57 *
58 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
59 * represented by pfn_mapped[].
60 */
61 unsigned long max_low_pfn_mapped;
62 unsigned long max_pfn_mapped;
63
64 #ifdef CONFIG_DMI
65 RESERVE_BRK(dmi_alloc, 65536);
66 #endif
67
68
69 unsigned long _brk_start = (unsigned long)__brk_base;
70 unsigned long _brk_end = (unsigned long)__brk_base;
71
72 struct boot_params boot_params;
73
74 /*
75 * These are the four main kernel memory regions, we put them into
76 * the resource tree so that kdump tools and other debugging tools
77 * recover it:
78 */
79
80 static struct resource rodata_resource = {
81 .name = "Kernel rodata",
82 .start = 0,
83 .end = 0,
84 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
85 };
86
87 static struct resource data_resource = {
88 .name = "Kernel data",
89 .start = 0,
90 .end = 0,
91 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
92 };
93
94 static struct resource code_resource = {
95 .name = "Kernel code",
96 .start = 0,
97 .end = 0,
98 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
99 };
100
101 static struct resource bss_resource = {
102 .name = "Kernel bss",
103 .start = 0,
104 .end = 0,
105 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
106 };
107
108
109 #ifdef CONFIG_X86_32
110 /* CPU data as detected by the assembly code in head_32.S */
111 struct cpuinfo_x86 new_cpu_data;
112
113 /* Common CPU data for all CPUs */
114 struct cpuinfo_x86 boot_cpu_data __read_mostly;
115 EXPORT_SYMBOL(boot_cpu_data);
116
117 unsigned int def_to_bigsmp;
118
119 struct apm_info apm_info;
120 EXPORT_SYMBOL(apm_info);
121
122 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
123 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
124 struct ist_info ist_info;
125 EXPORT_SYMBOL(ist_info);
126 #else
127 struct ist_info ist_info;
128 #endif
129
130 #else
131 struct cpuinfo_x86 boot_cpu_data __read_mostly;
132 EXPORT_SYMBOL(boot_cpu_data);
133 #endif
134
135
136 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
137 __visible unsigned long mmu_cr4_features __ro_after_init;
138 #else
139 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
140 #endif
141
142 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
143 int bootloader_type, bootloader_version;
144
145 /*
146 * Setup options
147 */
148 struct screen_info screen_info;
149 EXPORT_SYMBOL(screen_info);
150 struct edid_info edid_info;
151 EXPORT_SYMBOL_GPL(edid_info);
152
153 extern int root_mountflags;
154
155 unsigned long saved_video_mode;
156
157 #define RAMDISK_IMAGE_START_MASK 0x07FF
158 #define RAMDISK_PROMPT_FLAG 0x8000
159 #define RAMDISK_LOAD_FLAG 0x4000
160
161 static char __initdata command_line[COMMAND_LINE_SIZE];
162 #ifdef CONFIG_CMDLINE_BOOL
163 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
164 #endif
165
166 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
167 struct edd edd;
168 #ifdef CONFIG_EDD_MODULE
169 EXPORT_SYMBOL(edd);
170 #endif
171 /**
172 * copy_edd() - Copy the BIOS EDD information
173 * from boot_params into a safe place.
174 *
175 */
copy_edd(void)176 static inline void __init copy_edd(void)
177 {
178 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
179 sizeof(edd.mbr_signature));
180 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
181 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
182 edd.edd_info_nr = boot_params.eddbuf_entries;
183 }
184 #else
copy_edd(void)185 static inline void __init copy_edd(void)
186 {
187 }
188 #endif
189
extend_brk(size_t size,size_t align)190 void * __init extend_brk(size_t size, size_t align)
191 {
192 size_t mask = align - 1;
193 void *ret;
194
195 BUG_ON(_brk_start == 0);
196 BUG_ON(align & mask);
197
198 _brk_end = (_brk_end + mask) & ~mask;
199 BUG_ON((char *)(_brk_end + size) > __brk_limit);
200
201 ret = (void *)_brk_end;
202 _brk_end += size;
203
204 memset(ret, 0, size);
205
206 return ret;
207 }
208
209 #ifdef CONFIG_X86_32
cleanup_highmap(void)210 static void __init cleanup_highmap(void)
211 {
212 }
213 #endif
214
reserve_brk(void)215 static void __init reserve_brk(void)
216 {
217 if (_brk_end > _brk_start)
218 memblock_reserve(__pa_symbol(_brk_start),
219 _brk_end - _brk_start);
220
221 /* Mark brk area as locked down and no longer taking any
222 new allocations */
223 _brk_start = 0;
224 }
225
226 u64 relocated_ramdisk;
227
228 #ifdef CONFIG_BLK_DEV_INITRD
229
get_ramdisk_image(void)230 static u64 __init get_ramdisk_image(void)
231 {
232 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
233
234 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
235
236 if (ramdisk_image == 0)
237 ramdisk_image = phys_initrd_start;
238
239 return ramdisk_image;
240 }
get_ramdisk_size(void)241 static u64 __init get_ramdisk_size(void)
242 {
243 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
244
245 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
246
247 if (ramdisk_size == 0)
248 ramdisk_size = phys_initrd_size;
249
250 return ramdisk_size;
251 }
252
relocate_initrd(void)253 static void __init relocate_initrd(void)
254 {
255 /* Assume only end is not page aligned */
256 u64 ramdisk_image = get_ramdisk_image();
257 u64 ramdisk_size = get_ramdisk_size();
258 u64 area_size = PAGE_ALIGN(ramdisk_size);
259
260 /* We need to move the initrd down into directly mapped mem */
261 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
262 PFN_PHYS(max_pfn_mapped));
263 if (!relocated_ramdisk)
264 panic("Cannot find place for new RAMDISK of size %lld\n",
265 ramdisk_size);
266
267 initrd_start = relocated_ramdisk + PAGE_OFFSET;
268 initrd_end = initrd_start + ramdisk_size;
269 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
270 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
271
272 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
273
274 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
275 " [mem %#010llx-%#010llx]\n",
276 ramdisk_image, ramdisk_image + ramdisk_size - 1,
277 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
278 }
279
early_reserve_initrd(void)280 static void __init early_reserve_initrd(void)
281 {
282 /* Assume only end is not page aligned */
283 u64 ramdisk_image = get_ramdisk_image();
284 u64 ramdisk_size = get_ramdisk_size();
285 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
286
287 if (!boot_params.hdr.type_of_loader ||
288 !ramdisk_image || !ramdisk_size)
289 return; /* No initrd provided by bootloader */
290
291 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
292 }
293
reserve_initrd(void)294 static void __init reserve_initrd(void)
295 {
296 /* Assume only end is not page aligned */
297 u64 ramdisk_image = get_ramdisk_image();
298 u64 ramdisk_size = get_ramdisk_size();
299 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
300
301 if (!boot_params.hdr.type_of_loader ||
302 !ramdisk_image || !ramdisk_size)
303 return; /* No initrd provided by bootloader */
304
305 initrd_start = 0;
306
307 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
308 ramdisk_end - 1);
309
310 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
311 PFN_DOWN(ramdisk_end))) {
312 /* All are mapped, easy case */
313 initrd_start = ramdisk_image + PAGE_OFFSET;
314 initrd_end = initrd_start + ramdisk_size;
315 return;
316 }
317
318 relocate_initrd();
319
320 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
321 }
322
323 #else
early_reserve_initrd(void)324 static void __init early_reserve_initrd(void)
325 {
326 }
reserve_initrd(void)327 static void __init reserve_initrd(void)
328 {
329 }
330 #endif /* CONFIG_BLK_DEV_INITRD */
331
parse_setup_data(void)332 static void __init parse_setup_data(void)
333 {
334 struct setup_data *data;
335 u64 pa_data, pa_next;
336
337 pa_data = boot_params.hdr.setup_data;
338 while (pa_data) {
339 u32 data_len, data_type;
340
341 data = early_memremap(pa_data, sizeof(*data));
342 data_len = data->len + sizeof(struct setup_data);
343 data_type = data->type;
344 pa_next = data->next;
345 early_memunmap(data, sizeof(*data));
346
347 switch (data_type) {
348 case SETUP_E820_EXT:
349 e820__memory_setup_extended(pa_data, data_len);
350 break;
351 case SETUP_DTB:
352 add_dtb(pa_data);
353 break;
354 case SETUP_EFI:
355 parse_efi_setup(pa_data, data_len);
356 break;
357 default:
358 break;
359 }
360 pa_data = pa_next;
361 }
362 }
363
memblock_x86_reserve_range_setup_data(void)364 static void __init memblock_x86_reserve_range_setup_data(void)
365 {
366 struct setup_indirect *indirect;
367 struct setup_data *data;
368 u64 pa_data, pa_next;
369 u32 len;
370
371 pa_data = boot_params.hdr.setup_data;
372 while (pa_data) {
373 data = early_memremap(pa_data, sizeof(*data));
374 if (!data) {
375 pr_warn("setup: failed to memremap setup_data entry\n");
376 return;
377 }
378
379 len = sizeof(*data);
380 pa_next = data->next;
381
382 memblock_reserve(pa_data, sizeof(*data) + data->len);
383
384 if (data->type == SETUP_INDIRECT) {
385 len += data->len;
386 early_memunmap(data, sizeof(*data));
387 data = early_memremap(pa_data, len);
388 if (!data) {
389 pr_warn("setup: failed to memremap indirect setup_data\n");
390 return;
391 }
392
393 indirect = (struct setup_indirect *)data->data;
394
395 if (indirect->type != SETUP_INDIRECT)
396 memblock_reserve(indirect->addr, indirect->len);
397 }
398
399 pa_data = pa_next;
400 early_memunmap(data, len);
401 }
402 }
403
404 /*
405 * --------- Crashkernel reservation ------------------------------
406 */
407
408 #ifdef CONFIG_KEXEC_CORE
409
410 /* 16M alignment for crash kernel regions */
411 #define CRASH_ALIGN SZ_16M
412
413 /*
414 * Keep the crash kernel below this limit.
415 *
416 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
417 * due to mapping restrictions.
418 *
419 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
420 * the upper limit of system RAM in 4-level paging mode. Since the kdump
421 * jump could be from 5-level paging to 4-level paging, the jump will fail if
422 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
423 * no good way to detect the paging mode of the target kernel which will be
424 * loaded for dumping.
425 */
426 #ifdef CONFIG_X86_32
427 # define CRASH_ADDR_LOW_MAX SZ_512M
428 # define CRASH_ADDR_HIGH_MAX SZ_512M
429 #else
430 # define CRASH_ADDR_LOW_MAX SZ_4G
431 # define CRASH_ADDR_HIGH_MAX SZ_64T
432 #endif
433
reserve_crashkernel_low(void)434 static int __init reserve_crashkernel_low(void)
435 {
436 #ifdef CONFIG_X86_64
437 unsigned long long base, low_base = 0, low_size = 0;
438 unsigned long low_mem_limit;
439 int ret;
440
441 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
442
443 /* crashkernel=Y,low */
444 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
445 if (ret) {
446 /*
447 * two parts from kernel/dma/swiotlb.c:
448 * -swiotlb size: user-specified with swiotlb= or default.
449 *
450 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
451 * to 8M for other buffers that may need to stay low too. Also
452 * make sure we allocate enough extra low memory so that we
453 * don't run out of DMA buffers for 32-bit devices.
454 */
455 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
456 } else {
457 /* passed with crashkernel=0,low ? */
458 if (!low_size)
459 return 0;
460 }
461
462 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
463 if (!low_base) {
464 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
465 (unsigned long)(low_size >> 20));
466 return -ENOMEM;
467 }
468
469 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
470 (unsigned long)(low_size >> 20),
471 (unsigned long)(low_base >> 20),
472 (unsigned long)(low_mem_limit >> 20));
473
474 crashk_low_res.start = low_base;
475 crashk_low_res.end = low_base + low_size - 1;
476 insert_resource(&iomem_resource, &crashk_low_res);
477 #endif
478 return 0;
479 }
480
reserve_crashkernel(void)481 static void __init reserve_crashkernel(void)
482 {
483 unsigned long long crash_size, crash_base, total_mem;
484 bool high = false;
485 int ret;
486
487 total_mem = memblock_phys_mem_size();
488
489 /* crashkernel=XM */
490 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
491 if (ret != 0 || crash_size <= 0) {
492 /* crashkernel=X,high */
493 ret = parse_crashkernel_high(boot_command_line, total_mem,
494 &crash_size, &crash_base);
495 if (ret != 0 || crash_size <= 0)
496 return;
497 high = true;
498 }
499
500 if (xen_pv_domain()) {
501 pr_info("Ignoring crashkernel for a Xen PV domain\n");
502 return;
503 }
504
505 /* 0 means: find the address automatically */
506 if (!crash_base) {
507 /*
508 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
509 * crashkernel=x,high reserves memory over 4G, also allocates
510 * 256M extra low memory for DMA buffers and swiotlb.
511 * But the extra memory is not required for all machines.
512 * So try low memory first and fall back to high memory
513 * unless "crashkernel=size[KMG],high" is specified.
514 */
515 if (!high)
516 crash_base = memblock_phys_alloc_range(crash_size,
517 CRASH_ALIGN, CRASH_ALIGN,
518 CRASH_ADDR_LOW_MAX);
519 if (!crash_base)
520 crash_base = memblock_phys_alloc_range(crash_size,
521 CRASH_ALIGN, CRASH_ALIGN,
522 CRASH_ADDR_HIGH_MAX);
523 if (!crash_base) {
524 pr_info("crashkernel reservation failed - No suitable area found.\n");
525 return;
526 }
527 } else {
528 unsigned long long start;
529
530 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
531 crash_base + crash_size);
532 if (start != crash_base) {
533 pr_info("crashkernel reservation failed - memory is in use.\n");
534 return;
535 }
536 }
537
538 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
539 memblock_free(crash_base, crash_size);
540 return;
541 }
542
543 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
544 (unsigned long)(crash_size >> 20),
545 (unsigned long)(crash_base >> 20),
546 (unsigned long)(total_mem >> 20));
547
548 crashk_res.start = crash_base;
549 crashk_res.end = crash_base + crash_size - 1;
550 insert_resource(&iomem_resource, &crashk_res);
551 }
552 #else
reserve_crashkernel(void)553 static void __init reserve_crashkernel(void)
554 {
555 }
556 #endif
557
558 static struct resource standard_io_resources[] = {
559 { .name = "dma1", .start = 0x00, .end = 0x1f,
560 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
561 { .name = "pic1", .start = 0x20, .end = 0x21,
562 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
563 { .name = "timer0", .start = 0x40, .end = 0x43,
564 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
565 { .name = "timer1", .start = 0x50, .end = 0x53,
566 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
567 { .name = "keyboard", .start = 0x60, .end = 0x60,
568 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
569 { .name = "keyboard", .start = 0x64, .end = 0x64,
570 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
571 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
572 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
573 { .name = "pic2", .start = 0xa0, .end = 0xa1,
574 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
575 { .name = "dma2", .start = 0xc0, .end = 0xdf,
576 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
577 { .name = "fpu", .start = 0xf0, .end = 0xff,
578 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
579 };
580
reserve_standard_io_resources(void)581 void __init reserve_standard_io_resources(void)
582 {
583 int i;
584
585 /* request I/O space for devices used on all i[345]86 PCs */
586 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
587 request_resource(&ioport_resource, &standard_io_resources[i]);
588
589 }
590
snb_gfx_workaround_needed(void)591 static bool __init snb_gfx_workaround_needed(void)
592 {
593 #ifdef CONFIG_PCI
594 int i;
595 u16 vendor, devid;
596 static const __initconst u16 snb_ids[] = {
597 0x0102,
598 0x0112,
599 0x0122,
600 0x0106,
601 0x0116,
602 0x0126,
603 0x010a,
604 };
605
606 /* Assume no if something weird is going on with PCI */
607 if (!early_pci_allowed())
608 return false;
609
610 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
611 if (vendor != 0x8086)
612 return false;
613
614 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
615 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
616 if (devid == snb_ids[i])
617 return true;
618 #endif
619
620 return false;
621 }
622
623 /*
624 * Sandy Bridge graphics has trouble with certain ranges, exclude
625 * them from allocation.
626 */
trim_snb_memory(void)627 static void __init trim_snb_memory(void)
628 {
629 static const __initconst unsigned long bad_pages[] = {
630 0x20050000,
631 0x20110000,
632 0x20130000,
633 0x20138000,
634 0x40004000,
635 };
636 int i;
637
638 if (!snb_gfx_workaround_needed())
639 return;
640
641 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
642
643 /*
644 * SandyBridge integrated graphics devices have a bug that prevents
645 * them from accessing certain memory ranges, namely anything below
646 * 1M and in the pages listed in bad_pages[] above.
647 *
648 * To avoid these pages being ever accessed by SNB gfx devices reserve
649 * bad_pages that have not already been reserved at boot time.
650 * All memory below the 1 MB mark is anyway reserved later during
651 * setup_arch(), so there is no need to reserve it here.
652 */
653
654 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
655 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
656 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
657 bad_pages[i]);
658 }
659 }
660
trim_bios_range(void)661 static void __init trim_bios_range(void)
662 {
663 /*
664 * A special case is the first 4Kb of memory;
665 * This is a BIOS owned area, not kernel ram, but generally
666 * not listed as such in the E820 table.
667 *
668 * This typically reserves additional memory (64KiB by default)
669 * since some BIOSes are known to corrupt low memory. See the
670 * Kconfig help text for X86_RESERVE_LOW.
671 */
672 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
673
674 /*
675 * special case: Some BIOSes report the PC BIOS
676 * area (640Kb -> 1Mb) as RAM even though it is not.
677 * take them out.
678 */
679 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
680
681 e820__update_table(e820_table);
682 }
683
684 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)685 static void __init e820_add_kernel_range(void)
686 {
687 u64 start = __pa_symbol(_text);
688 u64 size = __pa_symbol(_end) - start;
689
690 /*
691 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
692 * attempt to fix it by adding the range. We may have a confused BIOS,
693 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
694 * exclude kernel range. If we really are running on top non-RAM,
695 * we will crash later anyways.
696 */
697 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
698 return;
699
700 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
701 e820__range_remove(start, size, E820_TYPE_RAM, 0);
702 e820__range_add(start, size, E820_TYPE_RAM);
703 }
704
early_reserve_memory(void)705 static void __init early_reserve_memory(void)
706 {
707 /*
708 * Reserve the memory occupied by the kernel between _text and
709 * __end_of_kernel_reserve symbols. Any kernel sections after the
710 * __end_of_kernel_reserve symbol must be explicitly reserved with a
711 * separate memblock_reserve() or they will be discarded.
712 */
713 memblock_reserve(__pa_symbol(_text),
714 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
715
716 /*
717 * The first 4Kb of memory is a BIOS owned area, but generally it is
718 * not listed as such in the E820 table.
719 *
720 * Reserve the first 64K of memory since some BIOSes are known to
721 * corrupt low memory. After the real mode trampoline is allocated the
722 * rest of the memory below 640k is reserved.
723 *
724 * In addition, make sure page 0 is always reserved because on
725 * systems with L1TF its contents can be leaked to user processes.
726 */
727 memblock_reserve(0, SZ_64K);
728
729 early_reserve_initrd();
730
731 memblock_x86_reserve_range_setup_data();
732
733 reserve_ibft_region();
734 reserve_bios_regions();
735 trim_snb_memory();
736 }
737
738 /*
739 * Dump out kernel offset information on panic.
740 */
741 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)742 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
743 {
744 if (kaslr_enabled()) {
745 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
746 kaslr_offset(),
747 __START_KERNEL,
748 __START_KERNEL_map,
749 MODULES_VADDR-1);
750 } else {
751 pr_emerg("Kernel Offset: disabled\n");
752 }
753
754 return 0;
755 }
756
757 /*
758 * Determine if we were loaded by an EFI loader. If so, then we have also been
759 * passed the efi memmap, systab, etc., so we should use these data structures
760 * for initialization. Note, the efi init code path is determined by the
761 * global efi_enabled. This allows the same kernel image to be used on existing
762 * systems (with a traditional BIOS) as well as on EFI systems.
763 */
764 /*
765 * setup_arch - architecture-specific boot-time initializations
766 *
767 * Note: On x86_64, fixmaps are ready for use even before this is called.
768 */
769
setup_arch(char ** cmdline_p)770 void __init setup_arch(char **cmdline_p)
771 {
772 #ifdef CONFIG_X86_32
773 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
774
775 /*
776 * copy kernel address range established so far and switch
777 * to the proper swapper page table
778 */
779 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
780 initial_page_table + KERNEL_PGD_BOUNDARY,
781 KERNEL_PGD_PTRS);
782
783 load_cr3(swapper_pg_dir);
784 /*
785 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
786 * a cr3 based tlb flush, so the following __flush_tlb_all()
787 * will not flush anything because the CPU quirk which clears
788 * X86_FEATURE_PGE has not been invoked yet. Though due to the
789 * load_cr3() above the TLB has been flushed already. The
790 * quirk is invoked before subsequent calls to __flush_tlb_all()
791 * so proper operation is guaranteed.
792 */
793 __flush_tlb_all();
794 #else
795 printk(KERN_INFO "Command line: %s\n", boot_command_line);
796 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
797 #endif
798
799 /*
800 * If we have OLPC OFW, we might end up relocating the fixmap due to
801 * reserve_top(), so do this before touching the ioremap area.
802 */
803 olpc_ofw_detect();
804
805 idt_setup_early_traps();
806 early_cpu_init();
807 jump_label_init();
808 static_call_init();
809 early_ioremap_init();
810
811 setup_olpc_ofw_pgd();
812
813 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
814 screen_info = boot_params.screen_info;
815 edid_info = boot_params.edid_info;
816 #ifdef CONFIG_X86_32
817 apm_info.bios = boot_params.apm_bios_info;
818 ist_info = boot_params.ist_info;
819 #endif
820 saved_video_mode = boot_params.hdr.vid_mode;
821 bootloader_type = boot_params.hdr.type_of_loader;
822 if ((bootloader_type >> 4) == 0xe) {
823 bootloader_type &= 0xf;
824 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
825 }
826 bootloader_version = bootloader_type & 0xf;
827 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
828
829 #ifdef CONFIG_BLK_DEV_RAM
830 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
831 #endif
832 #ifdef CONFIG_EFI
833 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
834 EFI32_LOADER_SIGNATURE, 4)) {
835 set_bit(EFI_BOOT, &efi.flags);
836 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
837 EFI64_LOADER_SIGNATURE, 4)) {
838 set_bit(EFI_BOOT, &efi.flags);
839 set_bit(EFI_64BIT, &efi.flags);
840 }
841 #endif
842
843 x86_init.oem.arch_setup();
844
845 /*
846 * Do some memory reservations *before* memory is added to memblock, so
847 * memblock allocations won't overwrite it.
848 *
849 * After this point, everything still needed from the boot loader or
850 * firmware or kernel text should be early reserved or marked not RAM in
851 * e820. All other memory is free game.
852 *
853 * This call needs to happen before e820__memory_setup() which calls the
854 * xen_memory_setup() on Xen dom0 which relies on the fact that those
855 * early reservations have happened already.
856 */
857 early_reserve_memory();
858
859 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
860 e820__memory_setup();
861 parse_setup_data();
862
863 copy_edd();
864
865 if (!boot_params.hdr.root_flags)
866 root_mountflags &= ~MS_RDONLY;
867 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
868
869 code_resource.start = __pa_symbol(_text);
870 code_resource.end = __pa_symbol(_etext)-1;
871 rodata_resource.start = __pa_symbol(__start_rodata);
872 rodata_resource.end = __pa_symbol(__end_rodata)-1;
873 data_resource.start = __pa_symbol(_sdata);
874 data_resource.end = __pa_symbol(_edata)-1;
875 bss_resource.start = __pa_symbol(__bss_start);
876 bss_resource.end = __pa_symbol(__bss_stop)-1;
877
878 #ifdef CONFIG_CMDLINE_BOOL
879 #ifdef CONFIG_CMDLINE_OVERRIDE
880 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
881 #else
882 if (builtin_cmdline[0]) {
883 /* append boot loader cmdline to builtin */
884 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
885 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
886 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
887 }
888 #endif
889 #endif
890
891 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
892 *cmdline_p = command_line;
893
894 /*
895 * x86_configure_nx() is called before parse_early_param() to detect
896 * whether hardware doesn't support NX (so that the early EHCI debug
897 * console setup can safely call set_fixmap()). It may then be called
898 * again from within noexec_setup() during parsing early parameters
899 * to honor the respective command line option.
900 */
901 x86_configure_nx();
902
903 parse_early_param();
904
905 if (efi_enabled(EFI_BOOT))
906 efi_memblock_x86_reserve_range();
907
908 #ifdef CONFIG_MEMORY_HOTPLUG
909 /*
910 * Memory used by the kernel cannot be hot-removed because Linux
911 * cannot migrate the kernel pages. When memory hotplug is
912 * enabled, we should prevent memblock from allocating memory
913 * for the kernel.
914 *
915 * ACPI SRAT records all hotpluggable memory ranges. But before
916 * SRAT is parsed, we don't know about it.
917 *
918 * The kernel image is loaded into memory at very early time. We
919 * cannot prevent this anyway. So on NUMA system, we set any
920 * node the kernel resides in as un-hotpluggable.
921 *
922 * Since on modern servers, one node could have double-digit
923 * gigabytes memory, we can assume the memory around the kernel
924 * image is also un-hotpluggable. So before SRAT is parsed, just
925 * allocate memory near the kernel image to try the best to keep
926 * the kernel away from hotpluggable memory.
927 */
928 if (movable_node_is_enabled())
929 memblock_set_bottom_up(true);
930 #endif
931
932 x86_report_nx();
933
934 if (acpi_mps_check()) {
935 #ifdef CONFIG_X86_LOCAL_APIC
936 disable_apic = 1;
937 #endif
938 setup_clear_cpu_cap(X86_FEATURE_APIC);
939 }
940
941 e820__reserve_setup_data();
942 e820__finish_early_params();
943
944 if (efi_enabled(EFI_BOOT))
945 efi_init();
946
947 dmi_setup();
948
949 /*
950 * VMware detection requires dmi to be available, so this
951 * needs to be done after dmi_setup(), for the boot CPU.
952 */
953 init_hypervisor_platform();
954
955 tsc_early_init();
956 x86_init.resources.probe_roms();
957
958 /* after parse_early_param, so could debug it */
959 insert_resource(&iomem_resource, &code_resource);
960 insert_resource(&iomem_resource, &rodata_resource);
961 insert_resource(&iomem_resource, &data_resource);
962 insert_resource(&iomem_resource, &bss_resource);
963
964 e820_add_kernel_range();
965 trim_bios_range();
966 #ifdef CONFIG_X86_32
967 if (ppro_with_ram_bug()) {
968 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
969 E820_TYPE_RESERVED);
970 e820__update_table(e820_table);
971 printk(KERN_INFO "fixed physical RAM map:\n");
972 e820__print_table("bad_ppro");
973 }
974 #else
975 early_gart_iommu_check();
976 #endif
977
978 /*
979 * partially used pages are not usable - thus
980 * we are rounding upwards:
981 */
982 max_pfn = e820__end_of_ram_pfn();
983
984 /* update e820 for memory not covered by WB MTRRs */
985 mtrr_bp_init();
986 if (mtrr_trim_uncached_memory(max_pfn))
987 max_pfn = e820__end_of_ram_pfn();
988
989 max_possible_pfn = max_pfn;
990
991 /*
992 * This call is required when the CPU does not support PAT. If
993 * mtrr_bp_init() invoked it already via pat_init() the call has no
994 * effect.
995 */
996 init_cache_modes();
997
998 /*
999 * Define random base addresses for memory sections after max_pfn is
1000 * defined and before each memory section base is used.
1001 */
1002 kernel_randomize_memory();
1003
1004 #ifdef CONFIG_X86_32
1005 /* max_low_pfn get updated here */
1006 find_low_pfn_range();
1007 #else
1008 check_x2apic();
1009
1010 /* How many end-of-memory variables you have, grandma! */
1011 /* need this before calling reserve_initrd */
1012 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1013 max_low_pfn = e820__end_of_low_ram_pfn();
1014 else
1015 max_low_pfn = max_pfn;
1016
1017 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1018 #endif
1019
1020 /*
1021 * Find and reserve possible boot-time SMP configuration:
1022 */
1023 find_smp_config();
1024
1025 early_alloc_pgt_buf();
1026
1027 /*
1028 * Need to conclude brk, before e820__memblock_setup()
1029 * it could use memblock_find_in_range, could overlap with
1030 * brk area.
1031 */
1032 reserve_brk();
1033
1034 cleanup_highmap();
1035
1036 memblock_set_current_limit(ISA_END_ADDRESS);
1037 e820__memblock_setup();
1038
1039 /*
1040 * Needs to run after memblock setup because it needs the physical
1041 * memory size.
1042 */
1043 sev_setup_arch();
1044
1045 efi_fake_memmap();
1046 efi_find_mirror();
1047 efi_esrt_init();
1048 efi_mokvar_table_init();
1049
1050 /*
1051 * The EFI specification says that boot service code won't be
1052 * called after ExitBootServices(). This is, in fact, a lie.
1053 */
1054 efi_reserve_boot_services();
1055
1056 /* preallocate 4k for mptable mpc */
1057 e820__memblock_alloc_reserved_mpc_new();
1058
1059 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1060 setup_bios_corruption_check();
1061 #endif
1062
1063 #ifdef CONFIG_X86_32
1064 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1065 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1066 #endif
1067
1068 /*
1069 * Find free memory for the real mode trampoline and place it there. If
1070 * there is not enough free memory under 1M, on EFI-enabled systems
1071 * there will be additional attempt to reclaim the memory for the real
1072 * mode trampoline at efi_free_boot_services().
1073 *
1074 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1075 * are known to corrupt low memory and several hundred kilobytes are not
1076 * worth complex detection what memory gets clobbered. Windows does the
1077 * same thing for very similar reasons.
1078 *
1079 * Moreover, on machines with SandyBridge graphics or in setups that use
1080 * crashkernel the entire 1M is reserved anyway.
1081 */
1082 reserve_real_mode();
1083
1084 init_mem_mapping();
1085
1086 idt_setup_early_pf();
1087
1088 /*
1089 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1090 * with the current CR4 value. This may not be necessary, but
1091 * auditing all the early-boot CR4 manipulation would be needed to
1092 * rule it out.
1093 *
1094 * Mask off features that don't work outside long mode (just
1095 * PCIDE for now).
1096 */
1097 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1098
1099 memblock_set_current_limit(get_max_mapped());
1100
1101 /*
1102 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1103 */
1104
1105 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1106 if (init_ohci1394_dma_early)
1107 init_ohci1394_dma_on_all_controllers();
1108 #endif
1109 /* Allocate bigger log buffer */
1110 setup_log_buf(1);
1111
1112 if (efi_enabled(EFI_BOOT)) {
1113 switch (boot_params.secure_boot) {
1114 case efi_secureboot_mode_disabled:
1115 pr_info("Secure boot disabled\n");
1116 break;
1117 case efi_secureboot_mode_enabled:
1118 pr_info("Secure boot enabled\n");
1119 break;
1120 default:
1121 pr_info("Secure boot could not be determined\n");
1122 break;
1123 }
1124 }
1125
1126 reserve_initrd();
1127
1128 acpi_table_upgrade();
1129 /* Look for ACPI tables and reserve memory occupied by them. */
1130 acpi_boot_table_init();
1131
1132 vsmp_init();
1133
1134 io_delay_init();
1135
1136 early_platform_quirks();
1137
1138 early_acpi_boot_init();
1139
1140 initmem_init();
1141 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1142
1143 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1144 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1145
1146 /*
1147 * Reserve memory for crash kernel after SRAT is parsed so that it
1148 * won't consume hotpluggable memory.
1149 */
1150 reserve_crashkernel();
1151
1152 memblock_find_dma_reserve();
1153
1154 if (!early_xdbc_setup_hardware())
1155 early_xdbc_register_console();
1156
1157 x86_init.paging.pagetable_init();
1158
1159 kasan_init();
1160
1161 /*
1162 * Sync back kernel address range.
1163 *
1164 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1165 * this call?
1166 */
1167 sync_initial_page_table();
1168
1169 tboot_probe();
1170
1171 map_vsyscall();
1172
1173 generic_apic_probe();
1174
1175 early_quirks();
1176
1177 /*
1178 * Read APIC and some other early information from ACPI tables.
1179 */
1180 acpi_boot_init();
1181 x86_dtb_init();
1182
1183 /*
1184 * get boot-time SMP configuration:
1185 */
1186 get_smp_config();
1187
1188 /*
1189 * Systems w/o ACPI and mptables might not have it mapped the local
1190 * APIC yet, but prefill_possible_map() might need to access it.
1191 */
1192 init_apic_mappings();
1193
1194 prefill_possible_map();
1195
1196 init_cpu_to_node();
1197 init_gi_nodes();
1198
1199 io_apic_init_mappings();
1200
1201 x86_init.hyper.guest_late_init();
1202
1203 e820__reserve_resources();
1204 e820__register_nosave_regions(max_pfn);
1205
1206 x86_init.resources.reserve_resources();
1207
1208 e820__setup_pci_gap();
1209
1210 #ifdef CONFIG_VT
1211 #if defined(CONFIG_VGA_CONSOLE)
1212 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1213 conswitchp = &vga_con;
1214 #endif
1215 #endif
1216 x86_init.oem.banner();
1217
1218 x86_init.timers.wallclock_init();
1219
1220 /*
1221 * This needs to run before setup_local_APIC() which soft-disables the
1222 * local APIC temporarily and that masks the thermal LVT interrupt,
1223 * leading to softlockups on machines which have configured SMI
1224 * interrupt delivery.
1225 */
1226 therm_lvt_init();
1227
1228 mcheck_init();
1229
1230 register_refined_jiffies(CLOCK_TICK_RATE);
1231
1232 #ifdef CONFIG_EFI
1233 if (efi_enabled(EFI_BOOT))
1234 efi_apply_memmap_quirks();
1235 #endif
1236
1237 unwind_init();
1238 }
1239
1240 #ifdef CONFIG_X86_32
1241
1242 static struct resource video_ram_resource = {
1243 .name = "Video RAM area",
1244 .start = 0xa0000,
1245 .end = 0xbffff,
1246 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1247 };
1248
i386_reserve_resources(void)1249 void __init i386_reserve_resources(void)
1250 {
1251 request_resource(&iomem_resource, &video_ram_resource);
1252 reserve_standard_io_resources();
1253 }
1254
1255 #endif /* CONFIG_X86_32 */
1256
1257 static struct notifier_block kernel_offset_notifier = {
1258 .notifier_call = dump_kernel_offset
1259 };
1260
register_kernel_offset_dumper(void)1261 static int __init register_kernel_offset_dumper(void)
1262 {
1263 atomic_notifier_chain_register(&panic_notifier_list,
1264 &kernel_offset_notifier);
1265 return 0;
1266 }
1267 __initcall(register_kernel_offset_dumper);
1268